[Haskell-cafe] Haskell language model

[Jorrit Willaert]

Dear all,

For my thesis, I will try to offer Haskell coding assignments for a course in a more dynamic way, where the next assignment is based on the knowledge gap of the learner. For this, I will use a neural network, which will have to process a submission of a learner. Therefore, I want to use a language model that extracts features from the submission.
I was planning to use BERT in a first phase, since this model creates embeddings from a document. However, this model is not trained on any programming language, so this may turn out to be insufficient. Moreover, models such as CodeBERT are trained on source code, but only on very well known programming languages.

I was wondering if someone had an idea if there already exists a language model that can create meaningful representations for Haskell. More general models such as Codex (paying API) or GPT-Code-Clippy (free and opensource) could also be tried out, but if there exists a model that has been trained on Haskell specific data, that would of course be the most beneficial option.

If you want to read more about my thesis, you can read more about it in the attachment.

Thanks in advance!

Kind regards,
Jorrit Willaert
https://www.linkedin.com/in/jorrit-willaert-166a64198/
https://github.com/JorritWillaert
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/haskell-cafe/attachments/20221019/6de5b2dd/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: masterproef.pdf
Type: application/pdf
Size: 257702 bytes
Desc: masterproef.pdf
URL: <http://mail.haskell.org/pipermail/haskell-cafe/attachments/20221019/6de5b2dd/attachment-0001.pdf>